[{"title":"The Rapid Evolution of AI Music Generation","url":"/posts/2026/06/ai-music-generation/","excerpt":"The pace at which generative models are advancing is nothing short of breathtaking. While much of the public discourse has been dominated by large language models (LLMs) and diffusion-based image generation, the frontier of generative...","content":"The pace at which generative models are advancing is nothing short of breathtaking. While much of the public discourse has been dominated by large language models (LLMs) and diffusion-based image generation, the frontier of generative audio has been quietly crossing a critical threshold: the synthesis of high-fidelity, emotionally resonant music.As a researcher, I spend most of my time analyzing the latent structures and theoretical limits of these models. However, occasionally, it is important to step back and simply appreciate the raw output.Recently, I experimented with state-of-the-art AI music generation tools. I provided the model with a few structural prompts, and..."},{"title":"Flight Simulators Are Becoming the AI Proving Ground","url":"/posts/2026/05/avion-level-d-ffs/","excerpt":"A Level D flight simulator is a legally certified digital twin.It is a massive computational and physical architecture costing tens of millions of dollars. Its engineering purpose is to flawlessly replicate the physical realities of...","content":"A Level D flight simulator is a legally certified digital twin.It is a massive computational and physical architecture costing tens of millions of dollars. Its engineering purpose is to flawlessly replicate the physical realities of flight. It recreates engine failures, zero-visibility landings, and hydraulic system collapses with such tightly certified precision that when a pilot steps out after two hours, EASA and the FAA legally sign off on them flying an Airbus A320 they have never actually flown.I have been at Avion since 2023 building these systems. But looking at the sheer volume of data flowing through these pipelines, a..."},{"title":"How to Recognize Advice That Actually Helps","url":"/posts/2025/09/recognize-helpful-advice/","excerpt":"Entrepreneurs, researchers, and engineers live in a torrent of guidance. Podcasts, newsletters, and mentors offer conflicting prescriptions, each delivered with confidence. Distinguishing signal from noise is therefore a crucial skill. The discipline of evidence based...","content":"Entrepreneurs, researchers, and engineers live in a torrent of guidance. Podcasts, newsletters, and mentors offer conflicting prescriptions, each delivered with confidence. Distinguishing signal from noise is therefore a crucial skill. The discipline of evidence based management argues that decisions should be grounded in the best available data rather than authority or habit (Pfeffer & Sutton, 2006). Applying this mindset to advice means scrutinizing both the source and the context before acting.First, effective advice answers a precisely formulated question. Psychologists have shown that vague prompts trigger confirmation bias, causing listeners to project their own assumptions onto the answer (Kahneman & Tversky,..."},{"title":"The Real Test for Making Something People Want","url":"/posts/2025/09/real-test-making-something-people-want/","excerpt":"Scaling conversations dominate startup culture, yet the first question any product must answer is painfully small: will even one person use it when given the chance? The only reliable way to find out is through...","content":"Scaling conversations dominate startup culture, yet the first question any product must answer is painfully small: will even one person use it when given the chance? The only reliable way to find out is through direct observation. This principle is at the heart of evidence based design, a methodology grounded in human computer interaction research that emphasizes empirical feedback over speculation (Nielsen, 1993). Surveys and interviews hint at preferences, but behavior reveals intent. If your prototype cannot hold the attention of a single user, no amount of marketing will redeem it.A prototype functions as an experimental apparatus. Eric Ries popularized..."},{"title":"Why Great Engineering Teams Run on Curiosity","url":"/posts/2025/09/startups-curiosity/","excerpt":"Working on simulator software, defensive cybersecurity platforms, and doctoral research has taught me that durable progress rarely starts with a perfect roadmap. It starts with persistent curiosity. Loewenstein’s review of the psychology of curiosity describes...","content":"Working on simulator software, defensive cybersecurity platforms, and doctoral research has taught me that durable progress rarely starts with a perfect roadmap. It starts with persistent curiosity. Loewenstein’s review of the psychology of curiosity describes it as a response to information gaps that push humans to seek missing data rather than settle for superficial explanations (Loewenstein, 1994). On the Avion telemetry team those gaps appeared as cockpit workflows that felt clumsy or metrics that refused to line up with pilot reports. Pulling on those loose threads, asking “why does this feel wrong?”, often exposed real product opportunities.Curiosity also provides a..."},{"title":"The Myth of the Quick Fix","url":"/posts/2025/09/myth-quick-fix/","excerpt":"Every engineer knows the temptation of the “five-minute hack.” A bug appears, the schedule is tight, and a clever shortcut promises to save the day. Weeks later, the team is still wrestling with the fallout....","content":"Every engineer knows the temptation of the “five-minute hack.” A bug appears, the schedule is tight, and a clever shortcut promises to save the day. Weeks later, the team is still wrestling with the fallout. The myth of the quick fix lies in its invisibility, short-term patches often masquerade as efficient solutions while secretly accruing interest that compounds over time.Bandaids Hide Systemic IssuesQuick fixes address symptoms, not root causes. When our Avion telemetry API began timing out under heavy load, we initially raised the request timeout from 30 seconds to 60 seconds. The issue vanished, until peak season, when latency..."},{"title":"Documentation as the First User Interface","url":"/posts/2025/09/documentation-first-ui/","excerpt":"Before a developer ever sees your landing page, they might encounter your README. For many technical products, documentation is the first user interface. It guides users through setup, communicates architecture decisions, and sets expectations for...","content":"Before a developer ever sees your landing page, they might encounter your README. For many technical products, documentation is the first user interface. It guides users through setup, communicates architecture decisions, and sets expectations for how the system behaves. Neglecting documentation is like shipping a GUI with missing buttons and unlabeled fields.Clear Docs Reduce Support LoadWell-written documentation prevents repetitive support requests and frees engineering time. When we shipped the first version of Avion’s telemetry API, the same three setup questions kept hitting our inbox, even though the answers technically lived in the docs. They were buried in long paragraphs without..."},{"title":"Hiring Generalists: The Small Team’s Secret Weapon","url":"/posts/2025/09/hiring-generalists/","excerpt":"The Case for Generalists In early-stage engineering, over-specialization is a hidden technical debt. I have observed that “Generalist” engineers act as the essential glue between disjointed systems, reducing the communication overhead that kills velocity. We...","content":"The Case for Generalists In early-stage engineering, over-specialization is a hidden technical debt. I have observed that “Generalist” engineers act as the essential glue between disjointed systems, reducing the communication overhead that kills velocity. We should hire for slope (learning rate) rather than just the y-intercept (current domain knowledge).Lean engineering teams face a paradox: they need specialists to build complex products, but they also need people who can switch contexts rapidly as priorities shift. Generalists thrive in this environment. They fill gaps no job description anticipated and create connective tissue between disciplines.Generalists Fill Invisible GapsWhen we built out Avion’s simulator..."},{"title":"The Invisible Cost of Shiny Tools","url":"/posts/2025/09/invisible-cost-shiny-tools/","excerpt":"Engineering teams love new tools. The promise of faster development, cleaner abstractions, or a more elegant stack is hard to resist. Yet every adoption decision carries hidden costs that rarely appear in procurement spreadsheets. The...","content":"Engineering teams love new tools. The promise of faster development, cleaner abstractions, or a more elegant stack is hard to resist. Yet every adoption decision carries hidden costs that rarely appear in procurement spreadsheets. The real price of a shiny tool lies in the cognitive load it introduces, the training it demands, and the fragile glue it creates between systems.Tool Churn Disrupts Team FlowSwitching tools interrupts the rhythm of a team. Documentation and institutional knowledge become outdated overnight. When we moved Comodo’s customer portal from a homegrown jQuery stack to React, three sprints vanished into retraining engineers and porting widgets..."},{"title":"METU Sailing Team: Racing in Urla, Bodrum, and Marmaris","url":"/posts/2025/09/metu-urla-marmaris-sailing/","excerpt":"Middle East Technical University’s sailing club has long been a hub for students who love the sea. I joined the club about six years ago, eager to escape Ankara’s dry inland climate, and spent nearly...","content":"Middle East Technical University’s sailing club has long been a hub for students who love the sea. I joined the club about six years ago, eager to escape Ankara’s dry inland climate, and spent nearly every free weekend on the Aegean. I’m not a member today, but the lessons still stick. Learning to call sail trim and feeling the hull surge after a clean tack gave me my first taste of real teamwork.Those early regattas taught me more than boat handling. We planned our own logistics, from borrowing a van for the sails to cooking pasta in the parking lot..."},{"title":"Why Early Beta Feedback Matters More Than Launch Day","url":"/posts/2025/09/early-beta-feedback/","excerpt":"Product launches often steal the spotlight, but the feedback gathered during early beta testing quietly determines whether launch day is a victory lap or a scramble to patch obvious gaps. Treating beta periods as a...","content":"Product launches often steal the spotlight, but the feedback gathered during early beta testing quietly determines whether launch day is a victory lap or a scramble to patch obvious gaps. Treating beta periods as a checkbox diminishes their power; when leveraged well, beta feedback drives the evolution of features, design, and infrastructure before they congeal into something hard to change.Early Feedback Uncovers Unknown UnknownsTeams come to beta tests with hypotheses about how users will interact with the product. Those hypotheses are rarely perfect. Real beta users explore workflows and edge cases developers never envisioned. While piloting Avion’s simulator diagnostics portal,..."},{"title":"Y Combinator and DNA: Fixed Points in Code and Biology","url":"/posts/2025/09/y-combinator-dna/","excerpt":"It started while I was nursing a Saturday-morning coffee and catching up on YouTube. Around the 11:20 mark of this talk, the speaker described the choreography of DNA replication, how strands of genetic code carry...","content":"It started while I was nursing a Saturday-morning coffee and catching up on YouTube. Around the 11:20 mark of this talk, the speaker described the choreography of DNA replication, how strands of genetic code carry instructions that produce the very enzymes, such as DNA polymerase, needed to copy themselves. The diagram on screen sparked an immediate flashback to the time I first wrangled the Y combinator into a Haskell lab assignment. Suddenly, the elegance of Church’s abstraction and the biology lecture overlapped in my head: both were telling a story about systems that bootstrap their own continuation.Back in that functional-programming..."},{"title":"Proof of Learning: Building Trust in Future Machine Learning","url":"/posts/2025/09/proof-of-learning-future/","excerpt":"Proof of Learning (PoL) verifies that a model was genuinely trained on claimed data by providing verifiable evidence of the training process. I first felt the urgency for PoL while helping an aerospace partner document...","content":"Proof of Learning (PoL) verifies that a model was genuinely trained on claimed data by providing verifiable evidence of the training process. I first felt the urgency for PoL while helping an aerospace partner document how a safety-critical model was trained, our counterparts were less interested in raw accuracy and more concerned about whether they could audit each gradient step. As machine learning systems become pervasive in critical domains, PoL offers a mechanism to ensure trust and accountability in model provenance.Why PoL Matters in Practice Model provenance: PoL links models to their training data and processes, deterring plagiarism and unauthorized..."},{"title":"Reflections on the PhD Journey: Building Secure AI in Availability Zones","url":"/posts/2025/09/phd-era-florida-ml-security/","excerpt":"Embarking on a PhD in Florida means waking up to salt air, sunscreen reminders, and the low rumble of aircraft testing engines at Embry-Riddle Aeronautical University. My advisor jokes that our campus smells like jet...","content":"Embarking on a PhD in Florida means waking up to salt air, sunscreen reminders, and the low rumble of aircraft testing engines at Embry-Riddle Aeronautical University. My advisor jokes that our campus smells like jet fuel and ambition; he’s right on both counts. By 7 a.m. I am usually rolling past the flight line with a notebook full of model checkpoints to validate and a thermos of Cuban coffee strong enough to double as rocket propellant.Morning Run-UpDaytona Beach may be famous for NASCAR and spring breakers, but my calendar reads more like the Level-D simulator test cards I drafted at..."},{"title":"Model Watermarking and the Future of Trustworthy AI","url":"/posts/2025/09/model-watermarking-future/","excerpt":"Model watermarking embeds identifiable patterns into a model’s parameters or outputs so that ownership can be demonstrated without access to the original training process. My work on trustworthy machine learning focuses on how ownership signals...","content":"Model watermarking embeds identifiable patterns into a model’s parameters or outputs so that ownership can be demonstrated without access to the original training process. My work on trustworthy machine learning focuses on how ownership signals can remain detectable after fine-tuning and transfer. Early work shows that deep network weights can carry hidden signatures without affecting accuracy [3], and behavior-based marks can survive aggressive fine-tuning [4,5].In production settings, watermark checks are only useful when paired with solid observability: low-latency telemetry ingest, reliable data paths, and verification hooks that can flag when a deployed model suddenly stops presenting the expected watermark. This..."},{"title":"ML-Enhanced Blockchain: Toward Intelligent, Adaptive Ledgers","url":"/posts/2025/09/ml-enhanced-blockchain/","excerpt":"Machine learning (ML) is increasingly used to make blockchain networks more secure, efficient, and user-friendly. When I co-authored our survey on blockchain-enhanced machine learning [1], I kept hearing founders ask the inverse question: “How can...","content":"Machine learning (ML) is increasingly used to make blockchain networks more secure, efficient, and user-friendly. When I co-authored our survey on blockchain-enhanced machine learning [1], I kept hearing founders ask the inverse question: “How can ML shore up my ledger right now?” This post collects the playbooks that stuck with me from those conversations and the lab prototypes we built afterward.Why I lean on ML for blockchains Security analytics: ML models can detect fraud, phishing, or money laundering patterns in real time, outperforming rule-based monitoring. Our incident response drills now include a lightweight classifier to flag unusual exchange flows before..."},{"title":"Blockchain Oracles and Smart Contracts: The World's Agile Board","url":"/posts/2025/09/blockchain-oracles-agile-board/","excerpt":"Traditional agile boards depend on centralized tools and manual updates. As projects span multiple organizations and trust boundaries, these boards struggle to provide real-time, trustworthy visibility. Blockchain oracles and smart contracts offer a path to...","content":"Traditional agile boards depend on centralized tools and manual updates. As projects span multiple organizations and trust boundaries, these boards struggle to provide real-time, trustworthy visibility. Blockchain oracles and smart contracts offer a path to an autonomous, tamper-evident coordination layer that could serve as the world’s agile board.Oracles bridge real-world progressOn-chain smart contracts cannot access external data directly. Oracles solve this by feeding verified off-chain events into the blockchain. Delivery confirmations, test results, or even IoT sensor readings can trigger contract state changes that reflect real-world progress. Modern oracle networks like Chainlink and UMA aggregate data from multiple independent nodes..."},{"title":"Blockchain-Enhanced Machine Learning: Linking Trust, Data, and Incentives","url":"/posts/2025/09/blockchain-enhanced-ml/","excerpt":"Editor’s Note on Distributed Intelligence Blockchain and Machine Learning are often discussed as separate revolutions, but their intersection offers a solution to the “black box” problem in AI. This article explores how we can move...","content":"Editor’s Note on Distributed Intelligence Blockchain and Machine Learning are often discussed as separate revolutions, but their intersection offers a solution to the “black box” problem in AI. This article explores how we can move model coordination on-chain while keeping heavy computation off-chain, a hybrid architecture that finally makes “Proof-of-Useful-Work” a reality.Blockchain technology and machine learning (ML) have traditionally evolved on parallel tracks. However, recent research shows that their convergence promises transparent, trustworthy, and decentralized intelligent systems. In our survey on blockchain-enhanced machine learning [1], we examine how the decentralized design of blockchain can address systemic challenges in ML and..."},{"title":"Autonomous Drones and Their Future Importance","url":"/posts/2025/09/autonomous-drones-future/","excerpt":"Consider this your two-minute hangar brief. Between February 2019 and November 2020, I served as an Expert Software Engineer at STM Defence Technologies in Ankara. My team and I wrote the operational brains for Kargu:...","content":"Consider this your two-minute hangar brief. Between February 2019 and November 2020, I served as an Expert Software Engineer at STM Defence Technologies in Ankara. My team and I wrote the operational brains for Kargu: a loitering munition that can patiently orbit before striking, and Togan: a reconnaissance UAV that refuses to land until it has found every point of interest. Coffee was strong, timelines were tighter than rotor tolerances, and every build had to keep aircrews safe.What “Mission Control” Actually MeansMission-control is where all the distributed systems magic happens. We carved the stack into micro-services long before it was..."},{"title":"USA vs Netherlands: An Expat's Perspective","url":"/posts/2024/06/usa-vs-nl-expat/","excerpt":"Living as an expat has given me the chance to compare daily life in the United States and the Netherlands in a way that guidebooks never could. My notebook is full of tiny contrasts: Leiden...","content":"Living as an expat has given me the chance to compare daily life in the United States and the Netherlands in a way that guidebooks never could. My notebook is full of tiny contrasts: Leiden grocery receipts, NS train schedules for my commute, and lab to-do lists from Daytona Beach.Work and Life BalanceIn Leiden, my mornings start with a bike ride toward the train that carries me to Avion Full Flight Simulators. Our engineering team builds flight-simulator software, and the culture is unapologetically focused on sustainable productivity. Code freezes really mean “log off,” and it is normal to see sprint..."},{"title":"Adversarial Examples for Proof of Learning","url":"/blog/adversarial-examples-for-proof-of-learning","excerpt":"Code for paper “Adversary examples” for Proof of LearningView on GitHub","content":"Code for paper “Adversary examples” for Proof of LearningView on GitHub"},{"title":"SecurePoL with Watermarking","url":"/blog/securepol-with-watermarking","excerpt":"Official code & experiments for my PhD dissertation: “Enhancing Security of Proof-of-Learning Against Spoofing Attacks Using Advanced Model Watermarking.”View on GitHub","content":"Official code & experiments for my PhD dissertation: “Enhancing Security of Proof-of-Learning Against Spoofing Attacks Using Advanced Model Watermarking.”View on GitHub"},{"title":"Image Measurement with PyQt5: A Cardiology Aid","url":"/blog/pyqt5_image_measurer","excerpt":"Python’s versatility allows it to create tools for a broad spectrum of professional fields. In this blog post, we delve into creating an image measurement tool using PyQt5 - a set of Python bindings for...","content":"Python’s versatility allows it to create tools for a broad spectrum of professional fields. In this blog post, we delve into creating an image measurement tool using PyQt5 - a set of Python bindings for Qt libraries. This tool, tailored for cardiology doctors analyzing EKG (Electrocardiogram) reports, lets users load an image, select two points on the image, and calculate the distance between them.This application, named “Image Measurer,” was developed for Dr. Görkem Şefik Fatihoğlu, aiding in his analysis of EKG reports.Project OverviewOur tool enables users to measure distances within an image relative to a real-world scale. For instance, users..."},{"title":"Automatic Detection of Cyber Security Events from Turkish Twitter Stream and Newspaper Data","url":"/posts/2021/02/icissp-presentation/","excerpt":"This presentation was about my paper titled “Automatic Detection of Cyber Security Events from Turkish Twitter Stream and Newspaper Data” at the International Conference on Information Systems Security and Privacy (ICISSP). Discussed the key findings...","content":"This presentation was about my paper titled “Automatic Detection of Cyber Security Events from Turkish Twitter Stream and Newspaper Data” at the International Conference on Information Systems Security and Privacy (ICISSP).  Discussed the key findings and insights from the paper,  Provided an overview of the research methodology and results,Participants were mostly researchers and practitioners interested in security and privacy challenges of information systems from both technological and social perspectives.The presentation is available at the provided link."},{"title":"MS Thesis","url":"/blog/ms-thesis","excerpt":"My Master of Science Thesis Project related researches, projects and documents.View on GitHub","content":"My Master of Science Thesis Project related researches, projects and documents.View on GitHub"},{"title":"Secure Proxy on Cloud Paper Demonstration","url":"/posts/2016/06/secure-proxy-on-cloud/","excerpt":"This demonstration was about my paper titled “Secure Proxy on Cloud”. Discussed the motivation behind creating secure boxes on the cloud, Explained the implementation details and challenges of the project, Shared short-term and long-term development...","content":"This demonstration was about my paper titled “Secure Proxy on Cloud”.  Discussed the motivation behind creating secure boxes on the cloud,  Explained the implementation details and challenges of the project,  Shared short-term and long-term development plans,  and surveyed related literature researches.Participants were mostly researchers, practitioners, and industry professionals interested in cloud-based security solutions.The demonstration is available at the provided link."},{"title":"Autonomous Cargo and Mail Delivery","url":"/publication/2014-clover-autonomous-delivery","excerpt":"Undergraduate project demonstrating autonomous cargo and mail delivery.","content":"Undergraduate project demonstrating autonomous cargo and mail delivery."},{"title":"Secure Proxy on Cloud","url":"/publication/SecureProxyCloud","excerpt":"This report presents a secure proxy solution for cloud-based services.Download paper here","content":"This report presents a secure proxy solution for cloud-based services.Download paper here"},{"title":"Automatic Detection of Cyber Security Events from Turkish Twitter Stream and Turkish Newspaper Data","url":"/publication/2019-metu-masters-thesis","excerpt":"Master’s thesis presenting automatic detection methods for cyber security events using Turkish Twitter stream and newspaper data.","content":"Master’s thesis presenting automatic detection methods for cyber security events using Turkish Twitter stream and newspaper data."},{"title":"Automatic Detection of Cyber Security Events from Turkish Twitter Stream and Newspaper Data","url":"/publication/AutomaticDetectionCyberSecurity","excerpt":"This paper presents an automatic detection method for cyber security events using Turkish Twitter stream and newspaper data.Download paper here","content":"This paper presents an automatic detection method for cyber security events using Turkish Twitter stream and newspaper data.Download paper here"},{"title":"Survey on Blockchain-Enhanced Machine Learning","url":"/publication/2023-ieee-access-survey","excerpt":"Survey of blockchain-enhanced machine learning approaches and systems.","content":"Survey of blockchain-enhanced machine learning approaches and systems."},{"title":"Enhancing Security of Proof-of-Learning against Spoofing Attacks using Feature-Based Model Watermarking","url":"/publication/2024-ieee-access-watermarking","excerpt":"Peer-reviewed article strengthening Proof-of-Learning security with feature-based watermarking.","content":"Peer-reviewed article strengthening Proof-of-Learning security with feature-based watermarking."},{"title":"Enhancing Proof-of-Learning Security Against Spoofing Attacks Using Model Watermarking","url":"/publication/2025-dissertation","excerpt":"Doctoral dissertation documenting novel methods to strengthen Proof-of-Learning against spoofing using model watermarking.","content":"Doctoral dissertation documenting novel methods to strengthen Proof-of-Learning against spoofing using model watermarking."},{"title":"SecurePoL: Integration of Watermarking With Proof-of-Learning to Enhance Security Against Spoofing Attacks","url":"/publication/2025-secureproofoflearning","excerpt":"SecurePoL presents a dual-layer framework that couples immutable Proof-of-Learning logs with three watermarking strategies (feature-based triggers, sparse parameter perturbations, and a non-intrusive auxiliary head), ensuring verification succeeds only when both the training trajectory and watermark...","content":"SecurePoL presents a dual-layer framework that couples immutable Proof-of-Learning logs with three watermarking strategies (feature-based triggers, sparse parameter perturbations, and a non-intrusive auxiliary head), ensuring verification succeeds only when both the training trajectory and watermark are consistent."},{"title":"Machine Learning & Mission-Critical Systems","url":"/","excerpt":"Trustworthy ML · Mission-Critical Systems · Engineering Leadership I design secure ML and mission-critical systems, establishing the standards that keep them correct when things fail. I'm Dr. Ozgur Ural, a senior software engineer and machine-learning...","content":"Trustworthy ML · Mission-Critical Systems · Engineering Leadership I design secure ML and mission-critical systems, establishing the standards that keep them correct when things fail. I'm Dr. Ozgur Ural, a senior software engineer and machine-learning researcher. Across Turkey, the United States, and the Netherlands, I've spent twelve years architecting systems that cannot be allowed to fail: real-time ground-control software for pioneering autonomous UAV programs, a national-scale data-leakage-prevention platform, and the real-time Level&nbsp;D flight-simulator platforms I build today at Avion. My Ph.D. in Machine Learning (ERAU, 2025) and my published research on proof-of-learning, model watermarking, and adversarial robustness give that engineering..."},{"title":"Technical Writing","url":"/blog/","excerpt":"Technical writing on ML security, distributed systems, and high-reliability software—written for researchers and engineering leaders. For longer-form parables, see the Essays; for interactive demos, visit the Lab. The Rapid Evolution of AI Music Generation 1...","content":"Technical writing on ML security, distributed systems, and high-reliability software—written for researchers and engineering leaders. For longer-form parables, see the Essays; for interactive demos, visit the Lab. The Rapid Evolution of AI Music Generation 1 minute read Published: June 17, 2026 Exploring the rapid advancements in generative audio and how sequence modeling has solved long-range dependencies in AI music generation. Flight Simulators Are Becoming the AI Proving Ground 7 minute read Published: May 19, 2026 What if flight simulators are the AI proving ground nobody is talking about? Three years inside Avion. How to Recognize Advice That Actually Helps 3..."},{"title":"Essays","url":"/essays/","excerpt":"Slow, indirect writing on the ideas behind my research — proof-of-learning, watermarking, adversarial machine learning, the quiet drift of distributed systems. Each piece is a small parable. They argue by example rather than by exposition;...","content":"Slow, indirect writing on the ideas behind my research — proof-of-learning, watermarking, adversarial machine learning, the quiet drift of distributed systems. Each piece is a small parable. They argue by example rather than by exposition; meaning is meant to gather behind the eyes after the page is closed. Read at the speed of a long walk. Approximately one essay every couple of months. Suggestions, corrections, or arguments are very welcome — open an issue or write me. Provenance · Machine learning The Watchmaker's Trial There is a story they tell in Antwerp, though no one will say where they heard..."},{"title":"Gradient Pinball: how machines learn, animated","url":"/lab/gradient-pinball/","excerpt":"← Back to Research Lab Deep Learning · Optimization ⛰️ Every modern model — GPT, diffusion, the autocomplete on your phone — learns the same way: it rolls a ball down a high-dimensional loss landscape,...","content":"← Back to Research Lab Deep Learning · Optimization ⛰️ Every modern model — GPT, diffusion, the autocomplete on your phone — learns the same way: it rolls a ball down a high-dimensional loss landscape, nudging billions of parameters in whichever direction makes the error smaller. This animation builds that picture from the gradient up — the learning-rate cliff, the momentum speedup, and the counter-intuitive truth about what actually traps an optimizer in a million dimensions. Scientific Reference: The optimization mechanism at the core of the deep neural networks studied in the author's machine-learning security research. Saddle-point prevalence follows Dauphin..."},{"title":"Universal Jira: Prediction Markets for Global Coordination, animated","url":"/lab/universal-jira/","excerpt":"← Back to Research Lab Mechanism Design × Blockchain ⮞ Centralized project management breaks down at global scale. But what if we replaced the Jira ticket with a Prediction Market? This animation shows the mechanism...","content":"← Back to Research Lab Mechanism Design × Blockchain ⮞ Centralized project management breaks down at global scale. But what if we replaced the Jira ticket with a Prediction Market? This animation shows the mechanism step by step: a bounty creates the market, the AMM turns trades into a price, and an oracle later resolves the outcome. The important quantity is not the ticket itself but the market's belief about whether the task will be completed. Scientific Reference: Based on mechanism design for decentralized autonomous organizations (DAOs), Automated Market Maker invariant math (e.g. Constant Product Market Makers), and Logarithmic Market..."},{"title":"ML Oracles: Bringing Truth to the Chain, animated","url":"/lab/oracles/","excerpt":"← Back to Research Lab Smart Contracts × Machine Learning Smart contracts are blind logic gates. They cannot see the outside world. An oracle feeds them data, but what if the data requires pattern recognition—like...","content":"← Back to Research Lab Smart Contracts × Machine Learning Smart contracts are blind logic gates. They cannot see the outside world. An oracle feeds them data, but what if the data requires pattern recognition—like analyzing a satellite image or classifying a loan application? You cannot run a neural network on-chain. This animation explains how zkML (Zero-Knowledge Machine Learning) and Optimistic Fraud Proofs bridge the gap, bringing off-chain AI inference on-chain with cryptographic certainty. Scientific Reference: Based on state-of-the-art research in decentralized oracle networks and verifiable machine learning, outlining the exact mechanisms by which a deterministic blockchain can trust a..."},{"title":"Proof-of-Learning (SecurePoL), animated","url":"/lab/training-fingerprint/","excerpt":"← Back to Research Lab Machine Learning · Model provenance 🔬 Anyone can download a model and claim they trained it — the final weights are just a tensor of numbers, copyable in milliseconds. The...","content":"← Back to Research Lab Machine Learning · Model provenance 🔬 Anyone can download a model and claim they trained it — the final weights are just a tensor of numbers, copyable in milliseconds. The proof is in the journey: a real training run leaves a checkpoint trajectory that took a full run to generate and is expensive to forge. This animation builds the idea from the ground up — why the path beats the point, how a verifier spot-checks it cheaply, the prover/adversary cost asymmetry, and how the author's SecurePoL seals the remaining crack with a watermark. Scientific Reference:..."},{"title":"Block Race: Nakamoto consensus, animated","url":"/lab/block-race/","excerpt":"← Back to Research Lab Blockchain · Nakamoto consensus ⛏️ Bitcoin's entire security argument is a foot race. Every ten minutes, every miner races to extend the chain; honest nodes always trust the longest one....","content":"← Back to Research Lab Blockchain · Nakamoto consensus ⛏️ Bitcoin's entire security argument is a foot race. Every ten minutes, every miner races to extend the chain; honest nodes always trust the longest one. To reverse a payment, an attacker must secretly outrun the whole network — and the probability they ever succeed decays exponentially in the number of confirmations you wait. This animation derives that probability from the ground up: the gambler's ruin, Satoshi's Poisson head-start, and the punchline that \"6 confirmations\" hides an assumption about who you're racing. Scientific Reference: The math is Nakamoto's whitepaper §11 (Calculations)...."},{"title":"Redundancy Reactor: fault tolerance, animated","url":"/lab/redundancy-reactor/","excerpt":"← Back to Research Lab Aerospace · Fault tolerance ✈️ Run three flight computers and a majority voter, and one faulty channel gets outvoted. Done right, redundancy turns the failure rate from q into qm+1...","content":"← Back to Research Lab Aerospace · Fault tolerance ✈️ Run three flight computers and a majority voter, and one faulty channel gets outvoted. Done right, redundancy turns the failure rate from q into qm+1 — superlinear safety. But \"three computers\" is only \"three independent failure paths\" if they fail differently. This animation derives the binomial-tail gain, then shows how a shared cause installs a floor ρq no amount of redundancy can beat — the exact mechanism that destroyed Ariane 5 in 1996. Scientific Reference: Triple Modular Redundancy with common-cause (β-factor) failure. The Ariane 5 Flight 501 case follows the..."},{"title":"The Watermarking Wars: Capacity vs. Robustness, animated","url":"/lab/watermarking-comparison/","excerpt":"← Back to Research Lab Machine Learning Security An evolution of the Model Heist Detector. Rather than examining a single Z-test, this animation explores the entire landscape of model watermarking strategies. We compare the mathematical...","content":"← Back to Research Lab Machine Learning Security An evolution of the Model Heist Detector. Rather than examining a single Z-test, this animation explores the entire landscape of model watermarking strategies. We compare the mathematical capacity and evasion robustness of white-box (sparse parameter perturbations), black-box (feature-based triggers), generative (LLM token bias), and architecture-level (non-intrusive auxiliary heads) watermarks. Scientific Reference: Synthesizing state-of-the-art watermarking capacity bounds and robustness theorems. Trigger set analysis follows Adi et al. (2018), generative watermarking follows Kirchenbauer et al. (2023). parameter perturbation (white-box) feature triggers (black-box) green-list tokens (LLMs) auxiliary head (architecture) scrubbing evasion attack ▸ What did..."},{"title":"Model Heist Detector: AI watermarks, animated","url":"/lab/model-heist/","excerpt":"← Back to Research Lab ML security · Model provenance 🕵️ Someone leaks your AI and fine-tunes it just enough to look different. Before you ever published, you spread a faint statistical signature across thousands...","content":"← Back to Research Lab ML security · Model provenance 🕵️ Someone leaks your AI and fine-tunes it just enough to look different. Before you ever published, you spread a faint statistical signature across thousands of weights — each mark too small to notice, but together a fingerprint only you can read. This animation shows why that works: one big mark is fragile, but k tiny correlated marks read back through a matched filter give detection power that climbs as √k — invisible in any one weight, undeniable across all of them. Scientific Reference: Implements robust statistical watermarking, modelled as..."},{"title":"Research Lab","url":"/lab/","excerpt":"Animated research Watch the mathematics move. Eight cinematic explainers built from my published research and the systems I engineer: Bitcoin consensus, AI watermarks, training-run forensics, fault tolerance, optimization, decentralized oracles, watermarking strategy comparison, and prediction...","content":"Animated research Watch the mathematics move. Eight cinematic explainers built from my published research and the systems I engineer: Bitcoin consensus, AI watermarks, training-run forensics, fault tolerance, optimization, decentralized oracles, watermarking strategy comparison, and prediction markets. Each one animates the real result from the equation up, written for a technical reader, neither dumbed down nor hand-waved. Every formula is exactly as it appears in the papers, including the approximations the papers themselves admit to. ▶ Start with my research: SecurePoL Read the papers behind the films 01 Block Race Why \"6 confirmations\" is a probability, not a promise: Bitcoin's §11..."},{"title":"Perspectives & Leadership","url":"/life/","excerpt":"Essays and notes on system architecture, driving technical excellence, the academia-to-industry transition, and what over a decade of shipping mission-critical software has taught me about building resilient systems. Written more loosely than the technical posts....","content":"Essays and notes on system architecture, driving technical excellence, the academia-to-industry transition, and what over a decade of shipping mission-critical software has taught me about building resilient systems. Written more loosely than the technical posts. METU Sailing Team: Racing in Urla, Bodrum, and Marmaris 3 minute read Published: September 10, 2025 Reflections on METU Sailing Club — racing on the Aegean coast in Urla, Bodrum, and Marmaris, and what competitive sailing taught me about teamwork. Reflections on the PhD Journey: Building Secure AI in Availability Zones 6 minute read Published: September 09, 2025 Embarking on a PhD in Florida means..."},{"title":"Projects","url":"/projects/","excerpt":"Selected work across Ph.D. research code, interactive demos, C++ systems primitives, and clinical tooling. Each project links to its repository and writeup. The Interactive Lab contains eight playable demos for the same research areas. Interactive...","content":"Selected work across Ph.D. research code, interactive demos, C++ systems primitives, and clinical tooling. Each project links to its repository and writeup. The Interactive Lab contains eight playable demos for the same research areas. Interactive · 5 experiments Research Lab: interactive experiments Eight interactive experiments covering distributed consensus, model watermarking, proof-of-learning, fault tolerance, and gradient descent. Each presents a real scenario with measurable outcomes. Built to make the core concepts of my research accessible in three minutes. Research Ph.D. · ML Security SecurePoL with Watermarking Reference implementation accompanying my dissertation. Integrates feature-based model watermarking with proof-of-learning to defend ML training..."},{"title":"Publications","url":"/publications/","excerpt":"Peer-reviewed research on trustworthy machine learning: proof-of-learning protocols, feature-based model watermarking, adversarial defences, and blockchain-enhanced ML. Three first-author IEEE Access papers (2023, 2024, 2025), one Ph.D. dissertation, and earlier work on NLP-driven cyber-security event detection....","content":"Peer-reviewed research on trustworthy machine learning: proof-of-learning protocols, feature-based model watermarking, adversarial defences, and blockchain-enhanced ML. Three first-author IEEE Access papers (2023, 2024, 2025), one Ph.D. dissertation, and earlier work on NLP-driven cyber-security event detection. Live record on Google Scholar. <h2>Journal Articles</h2><hr /> SecurePoL: Integration of Watermarking With Proof-of-Learning to Enhance Security Against Spoofing Attacks Published in IEEE Access, 2025 IEEE Access, December 2025. Recommended citation: Ural, O. and Yoshigoe, K. (2025). SecurePoL: Integration of Watermarking With Proof-of-Learning to Enhance Security Against Spoofing Attacks. IEEE Access, vol. 13, pp. 213067-213091. DOI: 10.1109/ACCESS.2025.3642198.Download Paper Enhancing Security of Proof-of-Learning against Spoofing Attacks..."},{"title":"Advisory & Service","url":"/services/","excerpt":"I advise engineering teams on trustworthy ML, model provenance, and high-reliability architecture. I also serve the research community as a reviewer and program committee member. Full publication record on Google Scholar. For engagements: drozgurural@gmail.com.Advisory &...","content":"I advise engineering teams on trustworthy ML, model provenance, and high-reliability architecture. I also serve the research community as a reviewer and program committee member. Full publication record on Google Scholar. For engagements: drozgurural@gmail.com.Advisory & speaking Advisory · ML security Trustworthy-ML & model-provenance review Independent assessment of training-integrity, watermarking, and model-provenance claims, grounded in my published proof-of-learning research. For teams shipping, buying, or auditing ML systems. Advisory · Architecture Mission-critical architecture review Design reviews for real-time and fault-tolerant systems, drawing on twelve years across defence, cybersecurity, and Level D flight simulation, including engineering leadership across complex products. Speaking Talks &..."},{"title":"CLAUDE.md","url":"/CLAUDE/","excerpt":"CLAUDE.mdThis file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.OverviewPersonal academic portfolio site for Dr. Ozgur Ural, built on a custom Jekyll template and hosted on GitHub Pages at ozgurural.github.io....","content":"CLAUDE.mdThis file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.OverviewPersonal academic portfolio site for Dr. Ozgur Ural, built on a custom Jekyll template and hosted on GitHub Pages at ozgurural.github.io. The site positions Ozgur for CTO/principal engineering roles — copy should lead with leadership and impact, not just academic credentials.Development CommandsNative Ruby (primary — Docker config is present but Ruby 3.3 is installed locally)bundle exec jekyll serve --port 4000 --force_polling # serve with polling (Windows)bundle exec jekyll build # build to _site/First build ~35 s, incremental rebuilds ~15–25 s. Browse at http://localhost:4000, not 127.0.0.1 —..."}]